Plain markdown files, versioned in git, is the format the most serious AI work is quietly moving to. If you are already writing this way, you are early, not weird. If you are not, this is the short read on why it is worth it.
memo/q3-strategy.md
The memo is a plain markdown file in a git repo.
Ship the redesign when it feels ready. Ship the redesign behind a flag on July 24.
An agent edit, caught in review.
Every tool you write in eventually goes away, gets acquired, changes its pricing, or stops opening your files. Plain text does not. A folder of markdown you can open in any editor, on any machine, in five years, without a login. That is the whole reason it wins.
Add git and you get history for free. Every change is saved. You can see what a document looked like last week, or a year ago. You can roll it back. Nothing is lost by accident.
If you have handed an AI a link to your Google Doc or your Notion page and watched it flail, this is why. Those tools wrap your words in menus, styles, tags, and scripts that a model has to read before it reaches the point.
Markdown drops all of that and keeps the parts an AI actually uses. The headings. The lists. The emphasis. It is, as one writer put it, “structured enough for machines and simple enough for humans.” The result feels like magic and is really just a cleaner input.
This is the part that surprised even me. In the last few months, three of the most watched people in AI landed on the same setup, for three different reasons.
“the real moat is shifting from models to files.”
Andrej Karpathy, a team at Google, and Garry Tan all reached for the same thing this year to make their AI remember: a folder of markdown, kept in git. Swap the model underneath and the work keeps running, because the value is in the files.
“Markdown is the new source code.”
The plan, the spec, the decisions get written in markdown. Whatever an AI produces from it is the disposable part. The document is what you keep and grow.
“structured enough for machines and simple enough for humans.”
A file a person can read at a glance and a model can parse without noise. Nothing else lives in that sweet spot.
Then you already know. The docs feel lighter. The AI feels sharper. Git catches your mistakes. You are not going back.
The one thing you gave up when you left Google Docs or Word: the review layer. Suggesting mode. Comments. Seeing a change before it becomes yours. In markdown that comes back as a diff in a pull request, which is fine for engineers and a wall for everyone else on your team.
Stet is the piece that was missing. It opens your plain markdown and puts the review layer back on top. An AI edit, or a teammate's pushed change, arrives as a proposal you can read in plain English. Accept, reject, or send it back with a note. The files stay yours. The history stays yours. You get the review back without giving up the format.
If markdown on git already feels obvious to you, Stet is the next obvious step.
You do not need to become a developer to try this. You do not need the command line. You need a folder on your machine, and a habit of saving your writing as .md files instead of docs locked inside a tool.
Start with one project. A blog post. A strategy memo. Your reading notes. Put the files in a folder. When you want an AI to help, point it at that folder and watch how much better it works when it can actually read what is there.
The version history is the part you will not know you needed until you have it. One day you will paste something bad over something good, and instead of panic, you will step back one version. That is the whole pitch for git, without any of the jargon.
Stet makes both parts feel like an editor, not a developer tool. Plain markdown, a proper writing surface, a review sidebar for anything AI touches, and history that quietly saves you when you need it.
And if you are wondering how Stet compares to the other tools living in this world, Obsidian, iA Writer, OpenKnowledge, and the rest, Where Stet fits is the map.